ARTICLE

Received 26 Sep 2014 | Accepted 22 May 2015 | Published 7 Jul 2015 DOI: 10.1038/ncomms8601 OPEN Prediction model for aneuploidy in early human embryo development revealed by single-cell analysis

Maria Vera-Rodriguez1,2,3, Shawn L. Chavez1,2,w, Carmen Rubio3,4, Renee A. Reijo Pera1,2,w & Carlos Simon2,3,4

Aneuploidies are prevalent in the human embryo and impair proper development, leading to cell cycle arrest. Recent advances in imaging and molecular and genetic analyses are postulated as promising strategies to unveil the mechanisms involved in aneuploidy generation. Here we combine time-lapse, complete chromosomal assessment and single-cell RT–qPCR to simultaneously obtain information from all cells that compose a human embryo until the approximately eight-cell stage (n ¼ 85). Our data indicate that the chromosomal status of aneuploid embryos (n ¼ 26), including those that are mosaic (n ¼ 3), correlates with significant differences in the duration of the first mitotic phase when compared with euploid embryos (n ¼ 28). Moreover, expression profiling suggests that a subset of is differentially expressed in aneuploid embryos during the first 30 h of development. Thus, we propose that the chromosomal fate of an embryo is likely determined as early as the pronuclear stage and may be predicted by a 12-gene transcriptomic signature.

1 Institute for Stem Cell Biology and Regenerative Medicine, Center for Reproductive and Stem Cell Biology, Stanford, California 94305, USA. 2 Department of Obstetrics and Gynecology, Institute for Stem Cell Biology and Regenerative Medicine, Center for Reproductive and Stem Cell Biology, Stanford, California 94305, USA. 3 IGenomix, Parc Cientific Universitat de Valencia, Catedratico Agustin Escardino 9, Paterna 46980, Valencia, Spain. 4 Department of Obstetrics and Gynecology, Fundacio´n Instituto Valenciano de Infertilidad, University of Valencia, INCLIVA Health Research Institute, Paterna 46980, Valencia, Spain. w Present addresses: Division of Reproductive and Developmental Sciences, Oregon National Primate Research Center, Beaverton, Oregon 97006, USA. Department of Obstetrics and Gynecology, Oregon Health and Science University, Beaverton, Oregon 97006, USA or Department Physiology and Pharmacology, Oregon Health and Science University, Beaverton, Oregon 97006, USA (S.L.C.). Department of Cell Biology and Neurosciences, Montana State University, Bozeman, Montana 59717, USA or Department of Chemistry and Biochemistry, Montana State University, Bozeman, Montana 59717, USA (R.A.R.P.). Correspondence and requests for materials should be addressed to C.S. (email: [email protected]).

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 1 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

ver the last 20 years, the number of assisted reproduction stages were measured for each embryo unless removed for procedures has drastically increased and is expected to molecular or chromosomal analysis before reaching this stage of Ocontinue to do so as parenthood is postponed. According development. We began by examining previously reported to European statistics, more than half a million in vitro parameters including the duration of the first cytokinesis, the fertilization (IVF) cycles are performed annually, resulting in time between the two- and three-cell, and time between the three- 100,000 newborns or 1.5% of all babies born in Europe1. and four-cell stages2,3 (Supplementary Table 1). The median Concomitant with the increased use of assisted reproduction, duration of the first cytokinesis was 20 min (range 15 min to there is an increase in the number of studies that have 2.9 h); the time from two- to three-cell stage was 11.4 h (range investigated the dynamics of human embryo development. 0–16.8 h); and the median time between the three- and four-cell Time-lapse imaging provides a non-invasive alternative to the stage was 1.3 h (range 0–22.1 h). Besides evaluating previously static morphological assessment of embryos, allowing the identified imaging parameters, we also measured the time evaluation of embryo viability via the measurement of between pronuclei disappearance (PNd) and the start of the predictive parameters on the basis of developmental kinetics2,3. first cytokinesis, a recently described parameter that has been In addition, time-lapse allows the observation of cellular events linked to human embryo viability4,5, but not to the chromosomal that would otherwise remain undetected by conventional status. As shown in Supplementary Table 1, the median for the methods such as bipolar one- to three-cell divisions4,5, time between PNd and the start of the first cytokinesis was 2.7 h fragment reabsorption6,7 and blastomere fusion5. Technologies (range 15 min to 22.2 h), suggesting that the wide range of such as array comparative genomic hybridization (aCGH) have this parameter might reflect underlying differences in embryo facilitated the transition from the study of nine developmental potential. using fluorescent in situ hybridization (FISH) to the analysis of all In addition to normal cell cycle divisions, we also were able to 23 pairs simultaneously in a single cell8. More detect abnormal division events in certain embryos, including recently, several studies have combined additional technologies in divisions from one to three cells or from one to four cells instead order to generate a ploidy prediction model on the basis of of the typical one to two cells. These atypical events were embryo kinetics7,9,10. observed in 23.6% (n ¼ 20) of embryos with the majority dividing Recent advances in the analysis of gene expression at the from one to three blastomeres (n ¼ 17; Supplementary Movie 1) single-cell level provide the opportunity to examine underlying and a much smaller subset dividing from one to four cells (n ¼ 3; molecular programmes involved in embryo ploidy generation. Supplementary Movie 2). Notably, 85% (n ¼ 17) of all abnormal Owing to technical limitations, previous studies of quantitative divisions occurred during the first mitosis, whereas only 15% reverse transcription PCR (RT–qPCR) of human embryo (n ¼ 3) occurred in either the second or third mitosis development analysed only a select group of genes11,12 and/or (Supplementary Movie 3), stressing the importance of the first large pools of embryos13, which can be confounded by potential mitotic division. Since irregular divisions produce a greater embryo heterogeneity. As RT–qPCR techniques have evolved and number of fast-dividing cells, the number of blastomeres was become more sensitive, gene expression analysis of individual disregarded as a method to stage the embryos. As an example, an human embryos followed3,14 and, more recently, single-cell embryo with four blastomeres typically results from three RT–qPCR analysis has become a reality3,15,16. To date, only a consecutive normal divisions; however, it could also be produced couple of studies have correlated gene expression patterns with during one direct cleavage from one to four cells. In both cases, aneuploidy in human embryos, one of which observed differential the embryos would be at the four-cell stage and it would be expression of certain epigenetic mediators in euploid versus inappropriate to similarly classify them. Thus, the time from PNd aneuploid embryos and the other examining DNA repair genes in was chosen as the starting reference point to define the embryo embryos with single over complex aneuploidies17,18. However, stage, since it was the first event observed post thaw and before the latter study did not include euploid embryos, only evaluated mitotic divisions. six chromosomes via FISH and analysed 15–20 pooled day-4 With regards to morphology, we also evaluated the incidence embryos rather than individual embryos or single cells. and timing of cellular fragmentation using multiplane imaging in Here we seek to examine the relationship between aneuploidy, order to obtain a three-dimensional model of each embryo and an embryo kinetics and transcriptome profiles at the earliest stages of accurate measurement of fragmentation degree. The vast majority the human preimplantation development. By simultaneously of embryos exhibited less than 25% fragmentation (n ¼ 62), assessing imaging behaviour, complete chromosomal composi- and no embryo showed fragmentation greater than 60% tion and the expression of B90 genes in single cells from whole (Supplementary Fig. 1a). Of the 64 embryos displaying fragmen- human embryos cultured at the one- to approximately eight-cell tation to any degree, most of them (68.8%) initially fragmented stage, we test the hypothesis that generation of aneuploidy may during the first division, 21.9% of embryos fragmented before and influence or be influenced by changes in gene expression and 9.4% after the completion of this division, further highlighting the embryo kinetics. Our results show that aneuploid embryos exhibit importance of the first mitosis (Supplementary Fig. 1b). altered developmental timing and a different transcriptomic profile than euploid embryos within the first 30 h of development. Thus, future studies should focus on the zygote as a key stage of Correlation between the ploidy status and developmental kinetics. human embryonic development and a potential source of non- Complete ploidy results were obtained for 89 cells from a total invasive biomarkers that can prospectively predict ploidy. of 57 embryos. The incidence of aneuploidy was 50.9% (n ¼ 29) and in agreement with previous reports7,19,20. In particular, 23 embryos were diagnosed with defined aneuploidies, 8 of them Results with a single chromosomal abnormality (Fig. 2a) and 15 with Early human embryo developmental kinetics and morphology. more than one chromosome affected. Interestingly, six embryos Eighty-five cryopreserved human zygotes were cultured for displayed mosaicism among blastomeres: in three embryos, all different time periods ranging from 2.3 to 64.9 h (Fig. 1). blastomeres were chromosomally abnormal but with different Developmental kinetics of each embryo were translated from abnormalities and complementary in some cases (Fig. 2b), frames to hours on the basis that an image frame was captured whereas three embryos exhibited a mixture of euploid and every 5 min. All timing intervals between the one- and nine-cell aneuploid blastomeres (Fig. 2c).

2 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

N = 117 Time-lapse imaging Human zygotes thawing

Embryo culture N = 85

Embryo Disassembly

Single-cell analysis

2.01 1.61 1.21 0.81 0.41 0.01 –0.40 –0.80 –1.20 –1.60 –2.00

1–0 7–5 9–9 2–60 5–47 21–5 3–127 11–41 13–82 16–67 Y–59 Quantitative PCR CGH arrays

Figure 1 | Experimental design of the study. One-hundred seventeen human embryos at the zygote stage were thawed, eighty-five of them survived and were cultured in nine different experiments. Embryo culture was performed in alphanumeric-labelled Petri dishes to allow embryo tracking during time-lapse imaging. Embryos were removed at different times until approximately the eight-cell stage. The number of cells varied depending on the type of divisions: one to two, one to three or one to four. Embryos were disaggregated into individual cells. Half of the cells from each embryo were analysed using aCGH to determine the ploidy status and the other half were analysed using RT–qPCR to study gene expression. Time-lapse movies were generated for each embryo and kinetic parameters were analysed.

We next evaluated developmental kinetics of aneuploid versus Mann–Whitney U-test), but not in embryos with abnormal euploid embryos to determine which parameter(s) may be divisions. Surprisingly, 4 of 20 embryos that underwent irregular correlated with ploidy. Only two imaging parameters were divisions also exhibited normal ploidy results. Three of these statistically significant between aneuploid and euploid embryos embryos were analysed immediately after the first mitosis—the when measured individually (Table 1). However, since embryos irregular one—and only one blastomere from each embryo was were taken out for analysis at different developmental times, it is analysed using aCGH. The fourth embryo also exhibited an important to note that the sample size gradually decreased as abnormal first division but was removed for analysis at the development proceeded, making the analysis of statistically nine-cell stage and the three blastomeres examined using aCGH significant differences for later parameters difficult. The most were male euploid. No multinucleation was detected in any of the significant parameter was the time between the PNd and the start four embryos and, notably, three of the four embryos were from of the first cytokinesis (P ¼ 0.025, Mann–Whitney U-test), which the same couple indicating an individual dimension of embryo was longer in aneuploid embryos compared with euploid development. embryos. As Table 1 demonstrates, the time between the three- We next determined whether there was a correlation between and four-cell stage was also statistically different between the embryo ploidy status and the incidence of fragmentation on aneuploid and euploid embryos (P ¼ 0.048, Mann–Whitney the basis of previous observations up to the four-cell stage7. For U-test) and approximately three times longer in aneuploid this purpose, we examined whether there was an association embryos. We also observed that the number of embryos with between aneuploidy and fragmentation degree in embryos with abnormal divisions in the aneuploid group was considerably low (o25%) versus high (Z25%) cellular fragmentation. While higher (12/26) when compared with the euploid group (4/22). 46.3% of the embryos with low fragmentation were aneuploid To determine whether there were differences in the ploidy status (19/41), the incidence in aneuploidy was 62.5% (10/16) in highly owing to this phenomenon, we analysed the developmental fragmented embryos. This observation was not statistically kinetics of normal versus abnormal embryo divisions separately significant and suggests that the measurement of fragmentation (Table 1), since abnormal divisions directly influence parameter degree alone is not predictive of the embryo ploidy status. timing. When we separated the embryos solely based on normal and abnormal divisions, we did not detect a significant difference in the timing from the three- to four-cell stage between euploid Assessment of single-blastomere gene expression profiles. and aneuploid embryos to confirm that abnormal divisions at this Single-cell gene expression results were obtained for 119 stage of development may reflect overall embryos’ ploidy status. blastomeres from 78 embryos. A total of 87 genes were selected Nevertheless, we detected statistically significant differences in the on the basis of their previously reported importance in the time between PNd and start of the first cytokinesis between literature3,21–35. The biological processes in which the genes were ploidy groups in embryos with normal divisions (P ¼ 0.049, involved included, but not limited to, cell cycle regulation,

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 3 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

a Aneuploid embryo

2.00 2.00 1.60 Blastomere 1 1.60 Blastomere 2 1.20 1.20 0.80 0.80 0.40 0.40 0.00 0.00 –0.40 –0.40 –0.80 –0.80 –1.20 –1.20 –1.60 –1.60 Log2 ratio Ch1/Ch2 Log2 ratio Log2 ratio Ch1/Ch2 Log2 ratio 1 2 3 4 5 6 7 8 9 10111213141516171819202122XY 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19202122 X Y Chromosomal position Chromosomal position

b Mosaic balanced aneuploid embryo

2.00 2.00 1.60 1.60 1.20 Blastomere 1 1.20 Blastomere 2 0.80 0.80 0.40 0.40 0.00 0.00 –0.40 –0.40 –0.80 –0.80 –1.20 –1.20 –1.60 –1.60 Log2 ratio Ch1/Ch2 Log2 ratio Log2 ratio Ch1/Ch2 Log2 ratio 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19202122 X Y 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19202122 X Y Chromosomal position Chromosomal position

c Mosaic euploid/aneuploid embryo

2.00 2.00 1.60 1.60 1.20 Blastomere 1 1.20 Blastomere 2 0.80 0.80 0.40 0.40 0.00 0.00 –0.40 –0.40 –0.80 –0.80 –1.20 –1.20 –1.60 –1.60 Log2 ratio Ch1/Ch2 Log2 ratio Ch1/Ch2 Log2 ratio

12345678910111213141516171819202122XY 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1718 19202122 X Y Chromosomal position Chromosomal position

2.00 2.00 1.60 Blastomere 3 1.60 Blastomere 4 1.20 1.20 0.80 0.80 0.40 0.40 0.00 0.00 –0.40 –0.40 –0.80 –0.80 –1.20 –1.20 –1.60 –1.60 Log2 ratio Ch1/Ch2 Log2 ratio Log2 ratio Ch1/Ch2 Log2 ratio

123456789101112131415161718 19202122 X Y 123456789101112131415161718 19202122 X Y Chromosomal position Chromosomal position

Figure 2 | Representative aCGH results. (a) Two blastomeres from the same four-cell embryo showing chromosome 17 trisomy. (b) Two blastomeres from a chromosomally mosaic four-cell embryo with balanced aneuploidies for chromosomes 2, 7, 16, 19 and the sex chromosomes (Y0 and XXY). (c) Four blastomeres from a mosaic eight-cell embryo with three euploid blastomeres (46 XY) and one blastomere with multiple aneuploidies. All profiles were compared with the male control DNA reference. apoptosis, telomere maintenance and DNA methylation. on the other hand, would increase with development and exhibit Individual gene expression patterns were analysed in individual no or relatively minor transcriptional inheritance from the blastomeres to determine the progression of expression levels gametes. during preimplantation development. To compare embryos at Cluster 1 (n ¼ 29) comprised genes inherited from the gametes different stages, PNd was designated as zero in the timescale and and showed no evidence of transcriptional activation by the 56 h afterwards was set as the final time point since no gene embryo since expression levels decreased throughout develop- expression data were obtained beyond this. In order to create a ment. On further analysis of the cluster 1 genes, we determined ‘best fit’ model that allowed the identification of statistically that the most significant annotations (Pr0.001, Fisher’s exact different gene profiles in embryos, a quadratic regression was test) were related to cell cycle regulation, DNA metabolism and performed for each gene as described in Methods. Using this chromosome organization (Table 2). In particular, aurora kinase regression, we observed statistically significant differences in 55 of A(AURKA), cadherin 1 (CDH1), cyclin-dependent kinase 7 the 87 genes analysed [analysis of variance (ANOVA) test, (CDK7), DNA methyltransferase 1 (DNMT1), peptidyl arginine Po0.05; Supplementary Table 2]. deiminase type VI (PADI6) and programmed cell death 5 We then grouped the genes into four different clusters (PDCD5) represented the major genes in this group (fold according to their expression dynamics (Fig. 3). For the change410 and Po1 Â 10 À 6, ANOVA test; Fig. 3). Our findings interpretation of the clusters, we considered that the gene are in accordance with previous reports3, wherein PDCD5, the expression values for each time point were the result of inherited cell death-related gene that inhibits the degradation of DNA molecules from the gametes and/or newly synthesized molecules damage response , was present in the zygote and generated by the embryonic genome. Thus, inherited transcripts decreased in expression until day-3. Moreover, AURKA, which would be highly expressed at the pronuclear stage and decrease in is involved in chromosome stabilization of the spindle, has also expression as development proceeds unless they are activated been shown to be highly expressed at early stages rather than by the embryonic genome. Embryonically transcribed genes, initially detected at the eight-cell stage21.

4 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

Table 1 | Kinetic parameters in euploid versus aneuploid embryos.

All embryos Normal divisions Abnormal divisions N Median IQR N Median IQR N Median IQR PNd to first cytokinesis (h) Euploid 22 2.4* (2.1; 2.9) 18 2.3z (2.1; 2.9) 4 2.9 (0.9; 3.1) Aneuploid 26 2.8* (2.5; 3.3) 14 2.8z (2.5; 3.3) 12 2.9 (2.5; 7.9)

First cytokinesis (min) Euploid 22 15.0 (13.8; 26.3) 18 15.0 (10.0; 21.3) 4 25.0 (16.3; 60.0) Aneuploid 26 20.0 (13.8; 31.3) 14 20.0 (10.0; 31.3) 12 22.5 (15.0; 37.5)

Two to three cells (h) Euploid 14 11.3 (1.4; 12.2) 10 11.7 (11.1; 12.6) 4 1.3 (0.5; 4.3) Aneuploid 23 11.4 (0.8; 12.6) 11 12.5 (11.8; 12.9) 12 0.8 (0.1; 2.4)

Three to four cells (h) Euploid 12 0.8w (0.2; 1.3) 10 0.8 (0.4; 1.3) 2 5.4 (0.1; NA) Aneuploid 20 2.4w (0.9; 8.4) 11 1.7 (0.8; 2.7) 9 4.4 (1.6; 12.3)

Four to five cells (h) Euploid 5 5.3 (1; 13.8) 4 8.8 (2.9; 14.6) 1 0 (0; 11.2) Aneuploid 25 9.9 (0.8; 12.3) 6 12.8 (11.7; 15.7) 9 2.7 (0.3; 9)

Five to six cells (h) Euploid 5 7.2 (1.4; 9.3) 4 4.9 (0.7; 7.4) 1 11.2 (11.2; 4.3) Aneuploid 12 1.6 (0.4; 3.8) 6 1.6 (0.5; 2.7) 6 1.6 (0.3; 13.8)

Six to seven cells (h) Euploid 4 1 (0.6; 3.5) 3 0.8 (0.5; NA) 1 4.3 (4.3; 0.8) Aneuploid 12 2 (1; 8.5) 6 1.5 (0.5; 3.9) 6 5.1 (1.2; 13.2)

Seven to eight cells (h) Euploid 4 0.8 (0.6; 1.1) 3 0.8 (0.5; NA) 1 0.8 (0.8; 0.5) Aneuploid 10 0.7 (0.4; 3.4) 6 1.5 (0.4; 3.4) 4 0.6 (0.3; 4.9)

Eight to nine cells (h) Euploid 1 0.5 (0.5; 0.5) 0 NA (NA; NA) 1 0.5 (0.5; NA) Aneuploid 2 2.5 (1.6; NA) 0 NA (NA; NA) 2 2.5 (1.6; NA)

IQR, interquartile range (Q1; Q3); NA, not applicable; PNd, pronuclei disappearance. Kinetic parameters were calculated for every aneuploid and euploid embryo (‘All embryos’). In addition, they were classified according to the type of divisions (‘Normal divisions’ and ‘Abnormal divisions’), since one abnormal division may alter all subsequent kinetic parameters, which are calculated based on the cell stage of the embryo. *,w,zPo0.05 (Mann–Whitney U-test).

Cluster 2 (n ¼ 4) was composed of genes that showed relatively in this cluster were Fas ligand (FASLG), growth arrest and constant expression and likely represent the bulk of transcripts DNA-damage-inducible alpha (GADD45A), SRY-box 2 (SOX2) inherited from the gametes since they were detected at the and zinc-finger and SCAN domain containing 4 (ZSCAN4; fold pronuclear stage, but also present at similar levels at later stages. change410 and Po1 Â 10 À 6, ANOVA test; Fig. 3). FASLG is a By avoiding mRNA degradation or if degraded, compensated for death receptor ligand whose expression has been shown to by new synthesis from the embryonic genome, these genes are correlate with fragmentation in human embryos at the two- and able to maintain stable levels throughout development. The genes four-cell stage37. We observed a gradual increase in FASLG that were statistically significant (Po0.05, ANOVA test) in this expression with development starting with basal levels at the cluster were v-akt murine thymoma viral oncogene homolog 1 pronuclear stage to suggest that embryos do not undergo (AKT1), breast cancer 1 (BRCA1), glyceraldehyde-3-phosphate apoptosis until later in development as previously described38. dehydrogenase (GAPDH) and NLR family, pyrin domain In addition, ZSCAN4, which is involved in telomere maintenance, containing 5 (NLRP5; Fig. 3). As a group, these genes have exhibited the greatest increase in expression of all the genes in known functions in monosaccharide metabolism and lipid this cluster (fold change ¼ 211), with considerably lower levels biosynthesis as well as food and stress responses or RNA stability before PNd. This was also in accordance with previous studies (Table 2). Notably, we detected high variability in BRCA1 showing ZSCAN4 expression in eight-cell embryos and no expression between cells from the same embryo, which could expression in zygotes30. explain the discordance with previous findings since these studies Cluster 4 (n ¼ 12) genes also increased in expression on detected differences between stages using the average expression embryonic genome activated (EGA), but, unlike Cluster 3, were of all equivalent samples24,36. also detected at the pronuclear stage to suggest both a gametic Cluster 3 (n ¼ 10) included genes that were activated during and embryo source of transcripts. Cyclin A1 (CCNA1), myeloid embryo development, but were not originally expressed in the cell leukaemia 1 (MCL1) and zygote arrest 1 (ZAR1) showed the zygote. analysis showed that these genes were most significant difference in this group (fold change410 and associated with regulation of the cell cycle, particularly interphase, Po1 Â 10 À 6, ANOVA test; Fig. 3). We detected low levels of as well as other biological processes such as the stress response CCNA1 expression at the pronuclear stage with significantly (Pr0.001, Fisher’s exact test; Table 2). The most relevant genes increased expression on EGA as previously described21 and

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 5 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

Expression Included Most significant pattern genes genes

AURKA MSH3 Cluster 1 AURKA CDH1 14 CDK7 DNMT1 AURKB NPM2 16 12 0 8 BUB3 OOEP 14 10 12 6 CDH1 PADI6 12 8 10 CDK7 PDCD5 4 10 6 8 2 Expression (log2) CHEK2 POT1 Expression (log2) Expression (log2) Expression (log2) 8 4 6 0 CRY1 RB1 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 CTNNB1 RCC2 Time from PNd (h) Time from PNd (h) Time from PNd (h) Time from PNd (h) DDX4 TERF2 PADI6 PDCD5 16 10 DNMT1 TP53 14 DNMT3A TSC2 12 8 DNMT3B TUBG1 10 6 E2F1 XPA 8 Expression (log2) Expression (log2) GATA4 YBX2 6 4 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 INCENP Time from PNd (h) Time from PNd (h)

Cluster 2

AKT1 BRCA1 NLRP5 5 16 GAPDH 6 AKT1 6 4 14 5 BRCA1 4 4 3 12 3 GAPDH 2 10 NLRP5 2 2 1 8 1 Expression (log2) Expression (log2) Expression (log2) Expression (log2) 0 0 6 0 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 Time from PNd (h) Time from PNd (h) Time from PNd (h) Time from PNd (h)

Cluster 3

FASLG GADD45A SOX2 ZSCAN4 10 2 BCL2 GADD45A 12.5 8 6 0 CASP2 MSH6 10.0 8 6 4 CCND1 RAD52 7.5 6 4 5.0 CDK2 SOX2 2 4 2 2.5 2 Expression (log2) Expression (log2) Expression (log2) Expression (log2) FASLG ZSCAN4 0 .0 0 0 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 –10 0 10 20 30 40 50 60 Time from PNd (h) Time from PNd (h) Time from PNd (h) Time from PNd (h)

Cluster 4

CCNA1 MCL1 ZAR1 ACTB MCL1 15 14 15 CCNA1 POU5F1 12 10 10 10 CDK1 RPLP0 8 CTCF RPS24 6 5 5 DDX20 YY1 4

Expression (log2) Expression (log2) 2 Expression (log2) DPPA3 ZAR1 0 0 0 –100 102030405060 –100 102030405060 –100 102030405060 Time from PNd (h) Time from PNd (h) Time from PNd (h)

Figure 3 | Identification of gene expression clusters during human embryo development. Significant quadratic regressions were classified into four different clusters according to the expression trend versus time. The majority of genes included in Cluster 1 (n ¼ 29) were expressed in the zygote stage but decreased in expression by at least twofold between the start and the final time. Cluster 2 (n ¼ 4) was composed of genes that showed relatively constant expression as defined by an increase or decrease in expression of less than 1 point. Cluster 3 (n ¼ 10) consisted of genes with an expression value lower than 2 at time zero and at least a twofold difference at the final time point. Lastly, Cluster 4 (n ¼ 12) comprised those genes with expression higher than 2 at time zero, and twofold or more at the last time point. The most significant regressions from each gene cluster were selected by ANOVA test with Po1 Â 10 À 6 and fold change410 for Clusters 1, 3, 4 and Po0.05 for Cluster 2. Each expression data point corresponds to the mean value obtained from three technical replicates. A baseline of Ct ¼ 28 was used to obtained the showed expression values. increasingly high levels of ZAR1 beginning at the pronuclear encode transcripts inherited from the gametes (Fig. 4a). stage. While CCNA1 binds particular cell cycle regulators, ZAR1 In this group, AURKA, BUB3 mitotic checkpoint (BUB3), is thought to function as a maternal effect gene in mouse and CDH1, CDK7, developmental pluripotency-associated 3 (DPPA3), human embryos39. However, gene ontology analysis did not show oocyte-expressed protein (OOEP) and PADI6 were the inherited any statistically significant annotations for this cluster. transcripts with the highest levels of expression at the zygote stage. In contrast, a total of 44 genes were clearly activated by the Identification and timing of gametic versus EGA transcripts. embryonic genome, 10 of which were undetectable in the zygote We next aimed to definitively determine which transcripts were and, thus, not likely required during the earliest stages of embryo from gametic origin or resulting from EGA. To accomplish this, development (Fig. 4a). we calculated the expression value at time zero for each gene For genes that showed clear activation on EGA (Clusters 3 using quadratic regression (Supplementary Table 2). Values and 4), we sought to determine the exact time in which activation higher than 2 indicated that the transcript was present in the occurs (Fig. 4b). The minimum of each quadratic function was zygote and thus provided by the gametes. When the expression calculated for this purpose and 8 of the 22 genes exhibited an values were above 2 for the final time point (Supplementary increase in expression starting from PNd. Interestingly, most of Table 2), the transcript was considered activated by EGA. the Cluster 3 genes were activated earlier in the development than From this analysis, we identified 40 genes that appeared to Cluster 4 genes. This can be explained by the finding that they

6 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

Table 2 | Gene ontology classifications for each cluster.

GO:Term Term name C G Adjusted P value Cluster 1 GO:0007049 Cell cycle 15 937 1.09E À 10 GO:0006259 DNA metabolic process 12 635 6.03E À 09 GO:0022402 Cell cycle process 11 638 1.00E À 07 GO:0009892 Negative regulation of metabolic process 11 707 2.20E À 07 GO:0009890 Negative regulation of biosynthetic process 10 561 3.68E À 07 GO:0051276 Chromosome organization 10 627 8.79E À 07 GO:0009893 Positive regulation of metabolic process 11 858 9.03E À 07 GO:0051726 Regulation of cell cycle 8 313 9.03E À 07 GO:0000278 Mitotic cell cycle 9 469 9.03E À 07 GO:0000279 M phase 8 386 3.88E À 06 GO:0009790 Embryonic development 9 608 6.72E À 06 GO:0033044 Regulation of chromosome organization 4 28 1.18E À 05 GO:0001701 In utero embryonic development 6 183 1.67E À 05 GO:0043009 Chordate embryonic development 7 344 2.63E À 05 GO:0051053 Negative regulation of DNA metabolic process 4 37 2.63E À 05 GO:0009792 Embryonic development ending in birth or egg hatching 7 348 2.64E À 05 GO:0010628 Positive regulation of gene expression 8 578 4.53E À 05 GO:0034984 Cellular response to DNA damage stimulus 7 382 4.53E À 05 GO:0050790 Regulation of catalytic activity 9 851 6.49E À 05 GO:0006974 Response to DNA damage stimulus 7 422 7.55E À 05 GO:0006366 Transcription from RNA polymerase II promoter 9 882 7.90E À 05 GO:0000087 M phase of mitotic cell cycle 6 269 8.95E À 05 GO:0016481 Negative regulation of transcription 7 450 1.00E À 04 GO:0016568 Chromatin modification 6 282 1.07E À 04 GO:0006461 Protein complex assembly 8 682 1.12E À 04 GO:0031328 Positive regulation of cellular biosynthetic process 8 700 1.25E À 04 GO:0006275 Regulation of DNA replication 4 65 1.25E À 04 GO:0010629 Negative regulation of gene expression 7 491 1.46E À 04 GO:0000075 Cell cycle checkpoint 4 77 2.16E À 04 GO:0008156 Negative regulation of DNA replication 3 26 4.36E À 04 GO:0051259 Protein oligomerization 5 217 4.91E À 04 GO:0051716 Cellular response to stimulus 8 865 4.99E À 04 GO:0033554 Cellular response to stress 7 623 5.80E À 04 GO:0032259 Methylation 4 106 6.07E À 04 GO:0006260 DNA replication 5 232 6.07E À 04 GO:0006306 DNA methylation 3 33 7.06E À 04 GO:0051096 Positive regulation of helicase activity 2 3 8.39E À 04 GO:0032206 Positive regulation of telomere maintenance 2 3 8.39E À 04 GO:0030521 Androgen receptor signalling pathway 3 37 8.96E À 04 GO:0007067 Mitosis 5 260 8.96E À 04 GO:0051128 Regulation of cellular component organization 6 458 9.51E À 04 GO:0043086 Negative regulation of catalytic activity 5 266 9.51E À 04

Cluster 2 GO:0006006 Glucose metabolic process 3 163 6.65E À 04 GO:0010907 Positive regulation of glucose metabolic process 2 12 6.65E À 04 GO:0019318 Hexose metabolic process 3 201 6.65E À 04 GO:0032094 Response to food 2 15 6.65E À 04 GO:0032369 Negative regulation of lipid transport 2 13 6.65E À 04 GO:0032770 Positive regulation of monooxygenase activity 2 17 6.65E À 04 GO:0034405 Response to fluid shear stress 2 8 6.65E À 04 GO:0045598 Regulation of fat cell differentiation 2 13 6.65E À 04 GO:0048009 Insulin-like growth factor receptor signalling pathway 2 16 6.65E À 04 GO:0050995 Negative regulation of lipid catabolic process 2 17 6.65E À 04 GO:0050999 Regulation of nitric-oxide synthase activity 2 13 6.65E À 04 GO:0051000 Positive regulation of nitric-oxide synthase activity 2 7 6.65E À 04 GO:0043487 Regulation of RNA stability 2 14 6.65E À 04 GO:0005996 Monosaccharide metabolic process 3 236 7.07E À 04 GO:0008633 Activation of pro-apoptotic gene products 2 19 7.08E À 04 GO:0043029 T-cell homeostasis 2 20 7.30E À 04 GO:0015909 Long-chain fatty acid transport 2 22 8.21E À 04 GO:0043491 Protein kinase B signalling cascade 2 23 8.43E À 04 GO:0046889 Positive regulation of lipid biosynthetic process 2 25 9.34E À 04 GO:0002260 Lymphocyte homeostasis 2 28 1.00E À 03 GO:0045862 Positive regulation of proteolysis 2 28 1.00E À 03 GO:0051353 Positive regulation of oxidoreductase activity 2 28 1.00E À 03

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 7 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

Table 2 (Continued )

GO:Term Term name C G Adjusted P value Cluster 3 GO:0022402 Cell cycle process 6 643 9.19E À 05 GO:0051325 Interphase 4 115 9.19E À 05 GO:0051329 Interphase of mitotic cell cycle 4 109 9.19E À 05 GO:0007049 Cell cycle 6 948 4.28E À 04 GO:0009411 Response to ultraviolet 3 55 7.31E À 04 GO:0033554 Cellular response to stress 5 625 9.57E À 04 GO:0033273 Response to vitamin 3 69 9.57E À 04 GO:0046661 Male sex differentiation 3 71 9.57E À 04

C, number of genes annotated by the given term in the test set; G, number of genes annotated by the given term in the reference set; GO, gene ontology. Two-tailed Fisher’s exact test was performed to detect most significant GO terms. Only results with Pr0.001 are shown. GO terms with Pr0.001 were not found for Cluster 4 genes. were not originally expressed in the gametes and thus needed to predictor. The Mann–Whitney U-test was performed between the be activated by the EGA at the earliest stage. In contrast, we aneuploid and euploid samples to obtain 31 differentially observed that the majority of the Cluster 4 genes did not need expressed genes (Po0.05). Before model construction, samples such early activation by the embryo, as there was already an initial were randomly split into a training group (n ¼ 27) and a pool of transcripts inherited from the zygote. validation group (n ¼ 14). Only the samples from the training group were used for the model design. To assess model accuracy, a fivefold cross-validation was performed and repeated 20 times Comparative gene expression analysis between ploidy groups. in order to estimate the misclassification rate. From this, several Ploidy and gene expression information was obtained from a total models were generated depending on the number of closest of 53 embryos with 92 cells analysed using RT–qPCR and 76 cells neighbours evaluated for the test step and the number of genes assessed via aCGH. To compare the expression profiles between selected for the training step. In order to obtain the most reliable ploidy groups without introducing bias due to differences in predictor, we applied additional restrictive parameters and only developmental time, two different groups were created using a selected genes with a P 0.005 (Mann–Whitney U-test, n ¼ 12). cutoff of 30 h after PNd. For the first group, we obtained o Cross-validation showed that the model with k ¼ 7 was the most 41 expression profiles from 33 embryos with an incidence of stringent [accuracy 85.2%, Matthews correlation coefficient aneuploidy of 39.4%; in the second group, we collected 52 (MCC) 0.62, root mean squared error (RMSE) 0.31, area under expression profiles from 20 embryos with a 75.0% aneuploidy the curve (AUC) 0.92]. Once selected, the predictor model was incidence. We determined that the incidence of aneuploidy was tested using the sample validation group. The confirmation rate higher in the second group because of the increased frequency of was 85.7% (12/14) with two of the euploid samples being mis- mitotic errors compared with the first group. We then used classified as aneuploid and no aneuploid samples inappropriately Babelomics40 to compare gene expression levels between the two called as euploid. Finally, we tested the prediction model in a groups. In the embryos collected before 30 h, 20 of the 87 different group of samples collected at the same time as the analysed genes showed statistically significant differences between training samples, but for which no ploidy results were obtained euploid and aneuploid embryos (adjusted P value 0.05, limma o (n ¼ 25). The prediction model classified 11 of the samples as test; Fig. 5a). We also evaluated the significant GO terms in these aneuploid and 14 as euploid with an incidence of aneuploidy of 20 genes to determine which molecular pathways were influenced 44.0%, which is similar to the observed rate in the samples with by aneuploid generation (Supplementary Table 3). We observed known ploidy results at the same stage (39.0%). We also com- that the most significant GO terms (n ¼ 16; P 0.001, Fisher’s o pared the time intervals between the PNd and the start of the first exact test) were related to cell cycle (7/16) and DNA damage cytokinesis in each embryo since this was the most relevant (5/16). More specifically, we determined that the expression of parameter observed for assessing ploidy status. As expected, the the catenin beta 1 (CTNNB1), Y box-binding protein 2 (YBX2) median time was much longer in the embryos classified as and tuberous sclerosis 2 (TSC2) were tremendously down- aneuploid versus those predicted to be euploid (2.58 versus regulated in aneuploid embryos compared with euploid 1.09 h). embryos (adjusted P value ¼ 0.01, limma test; Fig. 5b). In The predictor model identified 12 genes as applicable for contrast, the DNA damage response gene, GADD45A, was the classification of euploid versus aneuploid samples and highly expressed in aneuploid embryos and almost undetectable these included BUB1 mitotic checkpoint serine/threonine in euploid embryos (Fig. 5b). An analysis of gene expression 30 h kinase (BUB1), BUB3, caspase 2 (CASP2), CDK7, CTNNB1, E2F after PNd, however, did not show statistically significant transcription factor 1 (E2F1), GADD45A, GAPDH, pituitary differences between aneuploid and euploid embryos. tumour-transforming 1 (PTTG1), TP53, TSC2 and YBX2. With the exception of BUB1, CASP2, GAPDH and GADD45A, which Using transcriptomic signatures to predict embryo ploidy. were more highly expressed in aneuploid embryos, the majority Taking advantage of the difference in transcript expression of genes were upregulated in euploid embryos (Fig. 5a). Of the observed between euploid and aneuploid embryos during the first genes that exhibited lower expression in aneuploid embryos, 30 h of development, our next aim was to create a prediction many included maternally inherited genes such as BUB3, CDK7, model for embryo ploidy on the basis of a specific gene expression PTTG1, TSC2 and YBX2. Taken together, these data identify a signature (Fig. 6a). To accomplish this, cells from which both gene subset that is differentially regulated in euploid versus ploidy and gene expression data were obtained with a collection aneuploid embryos and suggests that mathematical modelling on time before 30 h post PNd were selected (n ¼ 41). Although the basis of the expression of this key group of genes may provide expression values from all 87 genes were available, we focused on a useful tool to largely predict the ploidy status of embryos the most informative genes to improve the functionality of the (Fig. 6b).

8 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

a Gametic and embryonic genes

CRY1 ACTB BCL2 AURKA AURKB CASP2 BUB3 CCNA1 CDH1 DNMT3B CCND1 CDK1 CDK7 CHEK2 CDK2 CTCF CTNNB1 DDX20 DDX4 GATA4 DNMT1 DNMT3A DPPA3 FASLG GAPDH INCENP MCL1 NPM2 GADD45A MSH3 OOEP PADI6 PDCD5 POT1 MSH6 POU5F1 RB1 RCC2 RPLP0 RAD52 XPA RPS24 TERF2 TP53 TUBG1 YY1 SOX2 YBX2 ZAR1 ZSCAN4

Gametic genes Embryonic genes

b Activated EGA genes SOX2 BCL2 CASP2 CCND1 FASLG GADD45A ZSCAN MSH6 RAD52 CDK2

01234567891011121314151617181920h

PNd

01234567891011121314151617181920h

Upregulated

EGA genes YY1 ZAR1 CTCF ACTB CDK1 MCL1 RPLP0 RPS24 DDX20 DPPA3 CCNA1 POU5F1

Figure 4 | Identification and timing of gametic versus embryonic transcripts. (a) Gametic transcripts (n ¼ 40) were highly expressed at time zero, whereas EGA genes (n ¼ 44) were highly expressed at the final time point. The majority of these genes (n ¼ 34) were originally inherited from the gametes and subsequently activated by EGA. (b) EGA timing is shown in hours (h) after PNd. Two groups were identified according to the basal levels at the pronuclear stage: ‘Activated EGA genes’ (n ¼ 10) that were originally absent at the zygote stage corresponded to Cluster 3 genes and ‘Upregulated EGA genes’ (n ¼ 12) that were already present at the pronuclear stage were designated for Cluster 4 genes. A schematic representation of embryo development with normal mitotic divisions was included as a guide.

Discussion abnormal. This further supports the notion that the high Significant advances in single-cell genetic profiling and time-lapse incidence of human embryonic aneuploidy at the cleavage stage imaging have dramatically increased the number of studies is often irrespective of the fertility status, maternal age and evaluating fundamental aspects of human preimplantation whether from fresh versus cryopreserved cycles. Regarding the development in the past 5 years4,5,16,18,25. Here we report the developmental kinetics of euploid versus aneuploid embryos, we first study to combine the analysis of complete chromosomal found only two parameters statistically different between both constitution, gene expression analysis and time-lapse culture groups. The most statistically significant parameter between simultaneously in the same human embryo. By evaluating all euploid and aneuploid embryos was the time between PNd and blastomeres from each embryo at a single-cell resolution across the start of the first cytokinesis. A recent study has shown this the first 3 days of development and correlating with embryo- parameter to be predictive of which embryos will reach the imaging behaviour, our results provide a mathematical model blastocyst stage5; however, no study has yet defined this predictive of ploidy status and a better understanding of early parameter for assessing the ploidy status. Here we determined human embryogenesis. that the time interval between PNd and the start of the first In accordance with previous findings using either FISH41,42 or cytokinesis was significantly longer in aneuploid than euploid different array-based approaches7,19,20, we observed that at least embryos, suggesting that chromosome missegregation can have half of the embryos in our cohort were indeed chromosomally an impact on the length of this first mitotic cycle. Thus, our

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 9 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

a Euploid Aneuploid Adjusted P-value GADD45A 1.0199e–2 GAPDH 1.4264e–2 BUB1 1.4897e–2 CASP2 1.4897e–2 CCND1 1.9139e–2 CCNA1 4.5895e–2 DDX20 4.5895e–2 AKT1 4.5895e–2 RPL10L 4.5895e–2 DNMT3B 4.5895e–2 NLRP5 4.5088e–2 PTTG1 4.5088e–2 ALKBH2 4.5088e–2 INCENP 4.4326e–2 CRY1 2.1342e–2 CDK7 1.4897e–2 E2F1 1.4264e–2 CTNNB1 1.0199e–2 YBX2 1.0199e–2 TSC2 1.0199e–2

b GADD45A CTNNB1 YBX2 TSC2 12 12 5 10 11 6 4 8 10 4 3 6 9 2 4 8 2 2 7 1 Expression (log2) Expression (log2) Expression (log2) Expression (log2) Expression 0 6 0 0 Euploid Aneuploid Euploid Aneuploid Euploid Aneuploid Euploid Aneuploid

Figure 5 | Differential gene expression in aneuploid versus euploid embryos. (a) Heatmap of genes showing significant differential expression (adjusted P valueo0.05, limma test; n ¼ 20) in euploid versus aneuploid embryos during the first 30 h after PNd. Each column represents a single blastomere. Blue coloured squares show low expression, while the red colour represents high levels of gene expression with white squares indicating moderate expression (log2). Expression data correspond to the mean values obtained from three technical replicates for each gene assay. (b) Box plots from the most significant differentially expressed genes (adjusted P value ¼ 0.01, limma test; n ¼ 4) between euploid and aneuploid embryos before 30 h following PNd. A plot represents gene expression values between quartile 1 and 3, the black line inside the box is the median value and the black circles are outliers.

a Gene expression samples b (N = 66) –4 420–2 Model generation Model test Comp.2 Comp.1 Samples with Samples without ploidy information ploidy information (N = 41) (N = 25) –4 .3

omp Informative gene selection –2 C Mann–Whitney U-test (P < 0.005) 0 BUB1 CTNNB1 PTTG1 BUB3 E2F1 TP53 CASP2 GADD45A TSC2 2 CDK7 GAPDH YBX2 1 0 Training set Validation set –1 (N = 27) (N = 14) –2 –3 KNN classification Confirmation rate Accuracy 85.2% 85.7% MCC 0.62 AUC 0.92

Cross-validation (K-fold method) RMSE 0.31

Figure 6 | Embryo ploidy prediction model. (a) A diagram of each phase of the ploidy prediction model. All samples with gene expression data were used in this process. Samples with ploidy results were selected for model generation and validation. Samples without ploidy information became the prediction group. MCC, Matthews correlation coefficient; AUC, area under the curve; RMSE, root mean squared error. (b) Principal component analysis of cells (n ¼ 41) from embryos at early stages (before 30 h post PNd) on the basis of the expression of the 12 genes selected in the prediction model. Cells from euploid embryos are shown in black (n ¼ 25) and samples from aneuploid embryos are designated red (n ¼ 16). results indicate that further consideration should be given to the Although there is some discrepancy as to whether time-lapse is competency of the sperm used for fertilization as is the oocyte actually beneficial for embryo selection45 and the assessment of since several mitotic spindle components for the first cell division ploidy status46, the first randomized control trial evaluating are paternally inherited in human embryos43,44. implantation rates, ongoing pregnancy and early pregnancy loss

10 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

Normal profile

Gene signature

BUB1 CTNNB1 PTTG1 BUB3 E2F1 TP53 CASP2 GADD45A TSC2 CDK7 GAPDH YBX2

Inherited transcriptome

Altered profile

Normal Altered Euploid Aneuploid Error during Long PNd to transcriptome transcriptome nucleus nucleus mitotic division first cytokinesis

Figure 7 | Proposed model of ploidy generation in early human embryo development. Aneuploidies in the human embryo are related to variations in both kinetics and expression profile. The inherited transcriptome at the zygote stage can be evaluated by a 12-gene signature to predict ploidy fate. When zygote transcripts levels are within normal ranges, embryo development occurs without errors in mitotic divisions and remain euploid. In contrast, when a zygote contains an altered transcriptome, mitotic errors will appear at any time throughout development. If the mitotic error happens before or during the first mitotic division, the time between PNd and the start of first cytokinesis will be longer than expected. If the mitotic error happens after the first mitotic division, on the other hand, the aneuploid embryo will not be detected by abnormal PNd to first cytokinesis kinetics but might be distinguishable by an altered transcriptome. suggests that dynamic imaging analysis is more effective than Since in the human embryo only a subset of genes is activated conventional IVF techniques9. It remains to be determined, before the eight-cell stage13,16,49,50, the first hours of development however, whether time-lapse imaging can also have positive rely on mRNA inherited from the gametes for survival. In this impact on live birth rates, particularly in cases of single-embryo study, we identified new potential maternally or paternally transfers. In addition, while the measurement of embryo kinetics derived gene products, including CDH1, which has been has been shown to differentiate a large proportion of aneuploid described as only expressed at the cleavage and blastocyst embryos from those that are euploid, chromosomally normal and stage25. Other genes, such as OOEP and PADI6, identified as abnormal embryos that behave similarly may be indistinguishable gametic in origin here have previously been reported as maternal and require genetic screening7,9,10,47. Nevertheless, it is clear from genes in murine embryos34, and our data suggest that they have a these studies that differences in findings may be because of the conserved function in early human development. This is stage of embryonic development evaluated, whether a cell of or particularly important as recent reports suggest distinct whole embryos were analysed, which method of ploidy differences in gene expression patterns between these two assessment was used and how the imaging parameters were species18,52,53. In accordance with recent findings16, we also measured since the use of a common start point such as the time observed the potential activation of a subset of genes involved of intracytoplasmic sperm injection may have confounding effects in cell cycle regulation, DNA metabolism or chromosome on other overlapping parameters2,46,48. We note that the time organization as early as the zygote stage. Further studies should between PNd and the first cytokinesis can be measured in both focus on the gametic source of these early-activated genes and non-cytoplasmic sperm injection and cryopreserved embryos, in their precise role in human preimplantation development. contrast to other parameters that may require the exact time of By combining single-cell gene expression and whole- fertilization as a reference point. chromosomal data analysis from the same embryo, we also Besides assessing the chromosomal status and imaging identified gene expression patterns indicative of the ploidy status. behaviour of each embryo, we also collected single-cell expression Notably, differential gene expression between aneuploid and data from 87 genes during the first stages of human embryo euploid embryos was observed during the first 30 h after PNd, but development and showed that 55 of them exhibit a defined not later. This lack of difference may be because of potential expression pattern. Although some of the genes have been overlap between maternal mRNA degradation and new tran- previously described in relation to human embryo viability, most scription by the embryonic genome, which can occur at diverse of these studies examined expression patterns in whole rates between embryos of the same developmental stage and even embryos14,24,30,36,37,49,50, in a pool of whole embryos13,17,21,26,27 in cells from the same embryo15. As uneven blastomere size is and/or only at a single stage of preimplantation associated with a high aneuploidy incidence54,55, asymmetric development17,21,36. In contrast, we evaluated gene expression distribution of transcripts caused by irregular divisions may also in single blastomeres throughout multiple stages of early contribute to this finding55,56. One of the key observations of our embryogenesis and observed a considerable amount of study is the discovery of a distinct set of genes, including CASP2, variability among samples. However, we addressed this CCND1, CCNA1, DDX20 and GADD45A, that are highly complication with the use of quantile normalization, a method expressed in aneuploid embryos. These genes further increase preferable to normalization to housekeeping genes because of in expression as development proceeds, which may be explained individual gene variation as previously described51. by the mitotic propagation of chromosomal errors that occurred

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 11 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 early in preimplantation development. We also determined that procedure were discarded and excluded from further analysis, since this could the majority of transcripts with decreased expression in aneuploid influence the integrity of RNA and DNA within the cells, thereby affecting the embryos belonged to Cluster 1 and, thus, likely of gametic origin. results. Embryos were cultured in custom polystyrene Petri dishes (Auxogyn, CA, USA) with 12 individual microwells in the centre. Small markers (letters and This group of genes included the cryptochrome circadian clock 1 numbers) were located at the edges to help with embryo identification. The dishes (CRY1), a circadian regulator protein that has been described to were prepared at least 5 h in advance and placed in the incubator to pre-equilibrate. 57 have a role in female meiosis , and the DNA methyltransferase The embryos were cultured at 37 °C with 6% CO2,5%O2 and 89% N2, standard 3B (DNMT3B). Furthermore, YBX2, the mouse homologue of human embryo culture conditions in accordance with current clinical IVF practice. MSY2, which binds maternal transcripts to avoid degradation during early embryo development58 and results in aberrant Time-lapse imaging. Embryos were monitored continuously using a microscope spindle formation when knocked out59, was also included in this system (Auxogyn) inside a standard tri-gas incubator (Sanyo, Japan). The system consisted of an inverted digital microscope with light-emitting diode illumination, group. Notably, both DNMT3B and YBX2 were previously shown  10 Olympus objective, automatic focus knob and 5 megapixel CMOS camera. to be expressed at significantly lower levels in human embryos Three types of images were taken during the culture: darkfield and brightfield that had arrested at the one- or two-cell stage3. Taken together, images were taken automatically every 5 min and at 1 s and 500 ms of exposure our findings suggest that the inheritance of an abnormal pool of time, respectively. In addition, brightfield images were also taken at 10 equidistant planes at several points throughout culture to capture images of the whole embryo. transcripts may contribute to aneuploidy in the human embryo The time between multiplane captures varied depending on when the embryo was (Fig. 7); however, it remains to be determined whether aberrant collected and one last capture was taken just before taking each embryo out of the gene expression is the potential cause or consequence of incubator. chromosomal errors in the gametes during meiosis. Finally, the most relevant finding of this study is that we were Embryo disassembly and collection. The embryos were collected at different able to largely predict embryo ploidy status using a 12-gene times and stages. For this purpose, the dish was taken out of the incubator for not transcriptomic signature. Although the power of the predictive more than 5 min to avoid affecting either the culture of the remaining embryos or the time-lapse imaging intervals. Embryos were individually transferred to 50 ml model has been determined and tested with embryos from several drops of Quinn’s Advantage Medium with HEPES (CopperSurgical) plus 10% clinics, it may be necessary to test our model on other embryo Human Albumin (HA; CooperSurgical) at 37 °C under mineral oil. Each procedure cohorts in order to extrapolate the results to other patient samples was performed with one single embryo at a time to maintain embryo identification and further substantiate our findings. Moreover, while it is not and tracking. The zona pellucida was removed from each embryo by transferring the embryos to 200 ml drops of Acidified Tyrode’s Solution (Millipore, MA, USA) currently intended for clinical use, the main goal of the prediction briefly and then washing in Quinn’s Advantage Medium plus 10% HA at 37 °C model is not to predict aneuploidy by itself, but to identify cellular under mineral oil. To weaken the cellular junctions between blastomeres, embryos pathways and related molecules indicative of the embryo were incubated in 60 ml drops of Quinn’s Advantage Ca þþ/Mg þþ-Free Medium ploidy status in culture medium or via other methods. Thus, with HEPES (CooperSurgical) plus 10% HA for 10 min at 37 °C under mineral oil. Embryos were disaggregated using gentle mechanical pipetting in the same we consider this study a keystone in the knowledge of early medium (Supplementary Fig. 2). human embryogenesis that may lead to the development of new Once disaggregated, counting and identification of blastomeres and polar non-invasive diagnostic tools that can reliably predict aneuploidy bodies was performed. Not all blastomeres from each embryo could be harvested. generation for IVF clinical routine. Annotations referring to the cell appearance such as visible nuclei, membrane integrity and cytoplasmic anomalies were recorded during the tubing. Each sample was washed three times in 5 ml drops of PBS 1% polyvinylpyrrolidone (PVP) buffer Methods and transferred to a sterile 0.2 ml PCR tube. All tubes were stored at À 80 °C until Experimental design. One-hundred seventeen human zygotes originating from subsequent analysis. 19 couples, with an average maternal age of 33.7±4.3 years, were thawed for this study. Eighty-five embryos survived and were cultured under time-lapse Time-lapse parameter evaluation. After each experiment, brightfield images imaging (Fig. 1), obtaining a survival rate of 72.6%, which is a normal value for were compiled into time-lapse movies using the ImageJ software66. Identification cryopreserved human embryos at the pronuclear stage60,61. Embryo retrieval was labels and time stamps were included to facilitate the measurement of the imaging performed at continuous times throughout embryonic development at the parameters. The imaging frames of several parameters were recorded, including pronuclear stage, and during one to seven mitotic divisions the number of the cells PNd, each division start time, beginning of the first cytokinesis, stage of varied depending on the following division types: one to two, one to three or one fragmentation appearance, final fragmentation percentage and the time in culture. to four cells. After embryo culture, embryos were disassembled into single Additional comments that could be useful for embryo evaluation such as the blastomeres, including polar bodies from zygotes. Half of the cells of each embryo presence of vacuoles, abnormal cell divisions and multinucleation were also underwent whole-genome amplification (WGA) and were analysed using aCGH recorded. In addition, multiplane images of every embryo were assembled in a to determine their chromosomal status at a single-cell level. The other half was multistack file using ImageJ. Multiplane captures were used to confirm brightfield analysed using real-time RT–qPCR for 87 genes to evaluate the specific and darkfield imaging observations and to assist in the measurement of certain transcriptome signature in each cell. Kinetic parameters, chromosomal status parameters such as PNd or percentage of fragmentation, which may be difficult to and expression levels were compared and analysed for individual embryos. determine using just one single plane. Embryo development evaluation was completed before ploidy and gene expression analyses to ensure blinded parameter Embryo source. For this study, we obtained a large set of human embryos from measurements. previous IVF cycles after written informed consent was obtained from the Stanford University RENEW Biobank. This cohort of embryos were cryopreserved at the Detection of chromosomal abnormalities in single cells. Single-cell DNA pronuclear stage before the assessment of quality, which have been shown to have extraction and WGA were accomplished using the Sureplex Kit (BlueGnome, 62,63 equivalent success rates as fresh sibling zygotes . It has also been recently Cambridge, UK) and tested using gel electrophoresis. WGA products and reference reported that there is no effect of oocyte cryopreservation on aneuploidy DNA (normal male and female controls) were labelled with either Cy3 or Cy5 64 65 incidence or kinetic parameters . Embryos in the RENEW Biobank are received fluorophores via the manufacturer’s instructions. Control and test-labelled DNAs from several IVF clinics across the United States. De-identification was performed were combined and co-hybridized on 24sure arrays (BlueGnome) for B12 h. After according to the Stanford University Institutional Review Board-approved protocol washing, slides were scanned using Innoscan 710 (Innopsys, Carbonne, France) no. 10466 entitled ‘The RENEW Biobank’, and molecular analysis of the embryos and the data analysed using the BlueFuse Multi software (BlueGnome). Results was in compliance with institutional regulations. No protected health information obtained from questionable samples, such as fragmented polar bodies or those with was associated with individual embryos. inconsistent aCGH profiles, were disregarded (Supplementary Fig. 3). Embryos without ploidy information were still useful to determine descriptive data about morphology, kinetics and gene expression in human embryos, as well as to train Embryo thawing and culture. Human embryos frozen at the two pronuclear the predictor model in a blind manner. stage by slow-freezing were thawed by a two-step process using the Quinn’s Advantage Thaw Kit (CooperSurgical, CT, USA) as recommended by the manufacturer. The embryos were washed in Quinn’s Advantage Cleavage Medium High-throughput single-cell gene expression analysis. Primers were designed (CooperSurgical) supplemented with 10% Quinn’s Advantage Serum Protein to span exons and detect all gene isoforms whenever possible (the primers are listed Substitute (CooperSurgical) and transferred to 100 ml drops of shared medium in Supplementary Table 4). The procedure for gene expression analysis was under mineral oil (Sigma, MO, USA). Embryos that did not survive the thaw adapted from the Advanced Development Protocol for Single-Cell Gene Expression

12 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601 ARTICLE

Using EvaGreen DNA Binding Dye (Fluidigm, CA, USA). In brief, cDNA was 4. Athayde Wirka, K. et al. Atypical embryo phenotypes identified by time-lapse prepared by adding to each individual sample the following: 9 ml RT-STA Solution microscopy: high prevalence and association with embryo development. Fertil. (5 ml Cells Direct 2 Â Reaction Mix (Invitrogen, CA, USA)); 0.2 ml SuperScript III Steril. 101, e1–e5 (2014). RT Platinum Taq Mix (Invitrogen); 2.5 ml 4X Primer Mix (200 nM); and 1.3 ml 5. Desai, N. et al. Analysis of embryo morphokinetics, multinucleation and DNA Suspension Buffer (Teknova, CA, USA). Reverse transcription and cleavage anomalies using continuous time-lapse monitoring in blastocyst pre-amplification were accomplished by incubating the samples at 50 °C for 15 min transfer cycles. Reprod. Biol. Endocrinol. 12, 12–54 (2014). ° ° ° and 95 C for 2 min, followed by 18 cycles of 95 C for 15 s and 60 C for 4 min. 6. Hardarson, T. et al. Internalization of cellular fragments in a human embryo: Exonuclease I treatment method was used to remove unincorporated primers and time-lapse recordings. Reprod. Biomed. 5, 36–38 (2002). the final volume was diluted twofold before qPCR. For qPCR, 2 ml of STA and Exo 7. Chavez, S. L. et al. Dynamic blastomere behaviour reflects human embryo I-treated sample was mixed with Sample Pre-Mix solution (2.5 ml2Â Taqman ploidy by the four-cell stage. Nat. Commun. 3, 1251 (2012). Gene Expression Master Mix (Applied Biosystems, CA, USA)); 20 Â DNA Binding Dye Sample Loading Reagent (Fluidigm); 20 Â EvaGreen DNA binding dye 8. Rodrigo, L. et al. New tools for embryo selection: comprehensive chromosome (Biotium, CA, USA). Gene assay mix solutions were prepared by adding 1.25 mlof screening by array comparative genomic hybridization. Biomed. Res. Int. 2014, 40 mM primer pairs with 2.5 ml2Â Assay Loading Reagent (Fluidigm) and 1.25 ml 517125 (2014). DNA Suspension Buffer. Sample and assay mixes were loaded into 96.96 Dynamic 9. Basile, N. et al. Increasing the probability of selecting chromosomally normal Arrays for qPCR on a BioMark System (Fluidigm). The same technical replicates embryos by time-lapse morphokinetics analysis. Fertil. Steril. 101, 699–704 were included on each dynamic array to check for variability between arrays and (2014). ensure reliable data. Data Collection and Real-Time PCR Analysis software 10. Campbell, A. et al. Modelling a risk classification of aneuploidy in human (Fluidigm) were used to calculate Ct values from the melt curve of each gene assay. embryos using non-invasive morphokinetics. Reprod. Biomed. 26, 477–485 (2013). 11. Taylor, D. M. et al. Quantitative measurement of transcript levels throughout Gene expression data-processing. Raw data were normalized in order to avoid human preimplantation development: analysis of hypoxanthine phosphoribosyl variability between chips and allow comparison between blastomeres from different transferase. Mol. Hum. Reprod. 7, 147–154 (2001). developmental stages. As gene activation during embryo development may not be 12. Tachataki, M., Winston, R. M. & Taylor, D. M. Quantitative RT-PCR reveals simultaneous in embryos of similar stage or between blastomeres within the same tuberous sclerosis gene, TSC2, mRNA degradation following cryopreservation embryo, normalization using housekeeping genes was not performed. Instead, in the human preimplantation embryo. Mol. Hum. Reprod. 9, 593–601 (2003). 67 68 a quantile normalization method was applied using limma package for R . 13. Zhang, P. et al. Transcriptome profiling of human pre-implantation This method adjusts the overall expression levels by making the distribution for all development. PLoS ONE 4, e7844 (2009). samples equal, thereby allowing us to compare expression values between cells 14. Shaw, L., Sneddon, S. F., Zeef, L., Kimber, S. J. & Brison, D. R. Global gene from different stages, or cells from the same embryo with different sizes. An expression profiling of individual human oocytes and embryos demonstrates assumed baseline Ct value of 28 or below was included on the basis of previous 69 heterogeneity in early development. PLoS ONE 8, e64192 (2013). findings and all Ct values higher than this value were called as no expression. 15. Yan, L. et al. Single-cell RNA-Seq profiling of human preimplantation embryos Similarly, all samples with questionable results such as a disproportionately and embryonic stem cells. Nat. Struct. Mol. Biol. 20, 1131–1139 (2013). high number of failed gene assays or unusual melt curves were discarded 16. Xue, Z. et al. Genetic programs in human and mouse early embryos revealed by (Supplementary Fig. 3). Final expression values were obtained by subtracting single-cell RNA sequencing. Nature 500, 593–597 (2013). C values from the C baseline value of 28. t t 17. Bazrgar, M. et al. DNA repair signalling pathway genes are overexpressed in poor-quality pre-implantation human embryos with complex aneuploidy. Eur. Statistical analysis of gene expression. Data analysis such as outliers, statistical J. Obstet. Gynecol. Reprod. Biol. 175, 152–156 (2014). tests and regression models were performed using IBM SPSS Statistics Version 21. 18. Chavez, S. L. et al. Comparison of epigenetic mediator expression and function Saphiro–Wilk test was first performed to check normal distribution of variables. If in mouse and human embryonic blastomeres. Hum. Mol. Genet. 23, 4970–4984 a normal distribution, t-test (for two groups) or one-way ANOVA (for more than (2014). two groups) was performed. When not normally distributed, the non-parametric 19. Vanneste, E. et al. Chromosome instability is common in human cleavage-stage Mann–Whitney U-test was performed to assess differences between groups. embryos. Nat. Med. 15, 577–583 (2009). For gene expression models we performed quadratic regression and ANOVA test 20. Johnson, D. S. et al. Preclinical validation of a microarray method for full was performed to evaluate the regression accuracy. Babelomics was the selected molecular karyotyping of blastomeres in a 24-h protocol. Hum. Reprod. 25, platform for the analysis of the gene expression data40. The functional analysis tool, 1066–1075 (2010). FatiGO, was used to detect over-represented functional annotations in a cluster of 21. Kiessling, A. A. et al. Genome-wide microarray evidence that 8-cell human genes and the class comparison tool assisted in the detection of genes differentially blastomeres over-express cell cycle drivers and under-express checkpoints. expressed between groups. Comparative expression between aneuploid and euploid J. Assist. Reprod. Genet. 27, 265–276 (2010). embryos from the same time point was performed in the class comparison tool 22. Mantikou, E., Wong, K. M., Repping, S. & Mastenbroek, S. Molecular origin of 40 using the limma test in Babelomics and Benjamini and Hochberg test was mitotic aneuploidies in preimplantation embryos. Biochim. Biophys. Acta 1822, selected to estimate the false discovery rate. 1921–1930 (2012). 23. Galan, A. et al. Defining the genomic signature of totipotency and pluripotency during early human development. PLoS ONE 8, e62135 (2013). Ploidy predictor model. The prediction model was built using the k-nearest 40 24. Wells, D. et al. Association of abnormal morphology and altered gene neighbours algorithm (k-NN) in the Babelomics platform , which consists of a expression in human preimplantation embryos. Fertil. Steril. 84, 343–355 function for the measurement of distances between samples on the basis of gene (2005). expression profiles. To avoid bias during model generation, sample-split was 25. Galan, A. et al. Functional genomics of 5- to 8-cell stage human embryos by performed. Two-thirds of the samples were randomly selected and grouped as a blastomere single-cell cDNA analysis. PLoS ONE 5, e13615 (2010). training set and the other one-third of the samples became the validation set, which was used to test the model once generated. Each sample from the training group 26. Jaroudi, S. et al. Expression profiling of DNA repair genes in human oocytes was assigned a class: euploid or aneuploid. For a test sample, the model was and blastocysts using microarrays. Hum. Reprod. 24, 2649–2655 (2009). assigned a class attending to the most represented among the closest k samples. 27. Jurisicova, A., Antenos, M., Varmuza, S., Tilly, J. L. & Casper, R. F. Expression Several models were generated on the basis of the number of neighbours that were of apoptosis-related genes during human preimplantation embryo evaluated for the prediction. In order to select the most accurate model, a k-fold development: potential roles for the Harakiri gene product and Caspase-3 in cross-validation was performed. In this method, the data set was automatically split blastomere fragmentation. Mol. Hum. Reprod. 9, 133–141 (2003). into k partitions and k–1 was used for model training and error estimation, 28. Liu, H. C. et al. Expression of apoptosis-related genes in human oocytes and respectively. This process was complete when all samples were tested and repeated embryos. J. Assist. Reprod. Genet. 17, 521–533 (2000). several times in order to improve prediction accuracy. 29. Jaroudi, S. & SenGupta, S. DNA repair in mammalian embryos. Mutat. Res. 635, 53–77 (2007). 30. Shaw, L., Sneddon, S. F., Brison, D. R. & Kimber, S. J. Comparison of gene References expression in fresh and frozen-thawed human preimplantation embryos. 1. Sullivan, E. A. et al. International Committee for Monitoring Assisted Reproduction 144, 569–582 (2012). Reproductive Technologies (ICMART) world report: assisted reproductive 31. Baran, V., Fabian, D. & Rehak, P. Akt/PKB plays role of apoptosis relay on technology 2004. Hum. Reprod. 28, 1375–1390 (2013). entry into first mitosis of mouse embryo. Zygote 21, 406–416 (2013). 2. Meseguer, M. et al. The use of morphokinetics as a predictor of embryo 32. Wan, L. B. et al. Maternal depletion of CTCF reveals multiple functions implantation. Hum. Reprod. 26, 2658–2671 (2011). during oocyte and preimplantation embryo development. Development 135, 3. Wong, C. C. et al. Non-invasive imaging of human embryos before embryonic 2729–2738 (2008). genome activation predicts development to the blastocyst stage. Nat. 33. Ma, J., Zeng, F., Schultz, R. M. & Tseng, H. Basonuclin: a novel mammalian Biotechnol. 28, 1115–1121 (2010). maternal-effect gene. Development 133, 2053–2062 (2006).

NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications 13 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8601

34. Tashiro, F. et al. Maternal-effect gene Ces5/Ooep/Moep19/Floped is 59. Medvedev, S., Pan, H. & Schultz, R. M. Absence of MSY2 in mouse oocytes essential for oocyte cytoplasmic lattice formation and embryonic perturbs oocyte growth and maturation, RNA stability, and the transcriptome. development at the maternal-zygotic stage transition. Genes Cells 15, Biol. Reprod. 85, 575–583 (2011). 813–828 (2010). 60. Reed, M. L., Hamic, A., Caperton, C. L. & Thompson, D. J. Live birth after 35. Bultman, S. J. et al. Maternal BRG1 regulates zygotic genome activation in the anonymous donation of twice-cryopreserved embryos that had been stored in mouse. Genes Dev. 20, 1744–1754 (2006). liquid nitrogen for a cumulative storage time of approximately 13.5 years. Fertil. 36. Giscard d’Estaing, S., Perrin, D., Lenoir, G. M., Guerin, J. F. & Dante, R. Steril. 94, e1–2771.e3 (2010). Upregulation of the BRCA1 gene in human germ cells and in preimplantation 61. Pavone, M. E., Innes, J., Hirshfeld-Cytron, J., Kazer, R. & Zhang, J. Comparing embryos. Fertil. Steril. 84, 785–788 (2005). thaw survival, implantation and live birth rates from cryopreserved zygotes, 37. Kawamura, K. et al. Expression of Fas and Fas ligand mRNA in rat and human embryos and blastocysts. J. Hum. Reprod. Sci. 4, 23–28 (2011). preimplantation embryos. Mol. Hum. Reprod. 7, 431–436 (2001). 62. Fugger, E. F. et al. Embryonic development and pregnancy from fresh and 38. Hardy, K. Apoptosis in the human embryo. Rev. Reprod. 4, 125–134 (1999). cryopreserved sibling pronucleate human zygotes. Fertil. Steril. 50, 273–278 39. Wu, X. et al. Zygote arrest 1 (Zar1) is a novel maternal-effect gene critical for (1988). the oocyte-to-embryo transition. Nat. Genet. 33, 187–191 (2003). 63. Miller, K. F. & Goldberg, J. M. In vitro development and implantation rates of 40. Medina, I. et al. Babelomics: an integrative platform for the analysis of fresh and cryopreserved sibling zygotes. Obstet. Gynecol. 85, 999–1002 (1995). transcriptomics, proteomics and genomic data with advanced functional 64. Goldman, K. N. et al. Long-term cryopreservation of human oocytes does not profiling. Nucleic Acids Res. 38, W210–W213 (2010). increase embryonic aneuploidy. Fertil. Steril. 103, 662–668 (2015). 41. Baart, E. B. et al. Preimplantation genetic screening reveals a high incidence of 65. De Munck, N. et al. Chromosomal meiotic segregation, embryonic aneuploidy and mosaicism in embryos from young women undergoing IVF. developmental kinetics and DNA (hydroxy)methylation analysis consolidate Hum. Reprod. 21, 223–233 (2006). the safety of human oocyte vitrification. Mol. Hum. Reprod. 21, 535–544 42. Munne, S. et al. Wide range of chromosome abnormalities in the embryos of (2015). young egg donors. Reprod. Biomed. Online 12, 340–346 (2006). 66. Schneider, C. A., Rasband, W. S. & Eliceiri, K. W. NIH Image to ImageJ: 43. Sathananthan, A. H. et al. Centrioles in the beginning of human development. 25 years of image analysis. Nat. Methods 9, 671–675 (2012). Proc. Natl Acad. Sci. USA. 88, 4806–4810 (1991). 67. Smyth, G. K. Limma: linear models for microarray data. Bioinformatics 44. Palermo, G., Munne, S. & Cohen, J. The human zygote inherits its mitotic Comput. Biol. Solut. 3, 397–420 (2005). potential from the male gamete. Hum. Reprod. 9, 1220–1225 (1994). 68. Core Team, R. in R: A language and Environment for Statistical Computing 45. Kaser, D. J. & Racowsky, C. Clinical outcomes following selection of human (R Foundation for Statistical Computing, 2013). preimplantation embryos with time-lapse monitoring: a systematic review. 69. Guo, G. et al. Resolution of cell fate decisions revealed by single-cell gene Hum. Reprod. 20, 617–631 (2014). expression analysis from zygote to blastocyst. Dev. Cell 18, 675–685 (2010). 46. Rienzi, L. et al. No evidence of association between blastocyst aneuploidy and morphokinetic assessment in a selected population of poor-prognosis patients: Acknowledgements a longitudinal cohort study. Reprod. Biomed. 30, 57–66 (2015). We gratefully acknowledge Gustavo Ayala and Nasser Al-Asmar for the assistance in the 47. Yang, Z. et al. Selection of competent blastocysts for transfer by combining aCGH experiments, Drs Juan M. Moreno-Moya, Antonia Dominguez and Mike Salt for time-lapse monitoring and array CGH testing for patients undergoing the help during the data normalization process, Dr Patricia Gomez-Diaz for many preimplantation genetic screening: a prospective study with sibling oocytes. helpful suggestions during the functional and statistical analysis for the model design and BMC Med.Genomics 7, 7–38 (2014). Martin Chian from Auxogyn Inc. for his assistance in the multicapture setting. We also 48. Dal Canto, M. et al. Cleavage kinetics analysis of human embryos predicts thank Dharti Trivedi who managed the Stanford Renew bank and all members of the development to blastocyst and implantation. Reprod. Biomed. 25, 474–480 Reijo Pera laboratory for technical assistance and invaluable discussions. This study was (2012). supported by the California Institute for Regenerative Medicine (Grant no. RB3-02209). 49. Vassena, R. et al. Waves of early transcriptional activation and pluripotency program initiation during human preimplantation development. Development Author contributions 138, 3699–3709 (2011). 50. Dobson, A. T. et al. The unique transcriptome through day 3 of human M.V.-R. designed and performed experiments, analysed data, wrote and edited the preimplantation development. Hum. Mol. Genet. 13, 1461–1470 (2004). manuscript. S.C.L. designed and performed experiments and wrote and edited the 51. Warren, L., Bryder, D., Weissman, I. L. & Quake, S. R. Transcription factor manuscript. C.R. designed experiments, interpreted results and assisted in the writing and editing of the manuscript. C.S. interpreted results and assisted in the writing and profiling in individual hematopoietic progenitors by digital RT-PCR. Proc. Natl editing of the manuscript. R.A.R.P. assisted in the design of experiments, interpreted Acad. Sci. USA 103, 17807–17812 (2006). results and assisted in the writing and editing of the manuscript. All the authors read and 52. Niakan, K. K. & Eggan, K. Analysis of human embryos from zygote to approved the manuscript. blastocyst reveals distinct gene expression patterns relative to the mouse. Dev. Biol. 375, 54–64 (2013). 53. Madissoon, E. et al. Differences in gene expression between mouse and Additional information human for dynamically regulated genes in early embryo. PLoS ONE 9, e102949 Supplementary Information accompanies this paper at http://www.nature.com/ (2014). naturecommunications 54. Magli, M. C., Gianaroli, L. & Ferraretti, A. P. Chromosomal abnormalities in embryos. Mol. Cell. Endocrinol. 183(): S29–S34 (2001). Competing financial interests: The authors declare no competing financial interests. 55. Hardarson, T., Hanson, C., Sjogren, A. & Lundin, K. Human embryos with Reprints and permission information is available online at http://npg.nature.com/ unevenly sized blastomeres have lower pregnancy and implantation rates: reprintsandpermissions/ indications for aneuploidy and multinucleation. Hum. Reprod. 16, 313–318 (2001). How to cite this article: Vera-Rodriguez, M. et al. Prediction model for aneuploidy in 56. Prados, F. J., Debrock, S., Lemmen, J. G. & Agerholm, I. The cleavage stage early human embryo development revealed by single-cell analysis. Nat. Commun. 6:7601 embryo. Hum. Reprod. 27, i50–i71 (2012). doi: 10.1038/ncomms8601 (2015). 57. Amano, T. et al. Expression and functional analyses of circadian genes in mouse oocytes and preimplantation embryos: Cry1 is involved in the meiotic This work is licensed under a Creative Commons Attribution 4.0 process independently of circadian clock regulation. Biol. Reprod. 80, 473–483 International License. The images or other third party material in this (2009). article are included in the article’s Creative Commons license, unless indicated otherwise 58. Yu, J., Hecht, N. B. & Schultz, R. M. Requirement for RNA-binding activity of in the credit line; if the material is not included under the Creative Commons license, MSY2 for cytoplasmic localization and retention in mouse oocytes. Dev. Biol. users will need to obtain permission from the license holder to reproduce the material. 255, 249–262 (2003). To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

14 NATURE COMMUNICATIONS | 6:7601 | DOI: 10.1038/ncomms8601 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved.